Developers

Build with trust from day one.

Integrate identity, authorization and evidence into your AI agent workflows. Three calls, one contract, no framework lock-in.

Developers

Build with trust from day one.

Integrate identity, authorization and evidence into your AI agent workflows with a single HTTP call before each action.

  • Authorization endpoint with recorded decision inputs
  • Action recording with integrity digests
  • Approval state you can poll or receive on review
  • Audit export in CSV and JSON
authorize-action
// Request authorization before the agent acts
const res = await fetch("https://api.certaion.com/v1/authorize", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.CERTAION_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    agent_id: "agt_1904",
    action: "refund.process",
    resource: "order_88213",
    context: { amount: 74.0, currency: "USD" },
  }),
});

const decision = await res.json();
// { effect: "require_approval", policy: "pol_refund", version: 3, approval_id: "apr_..." }

if (decision.effect === "allow") await processRefund();
else if (decision.effect === "require_approval") await queueForApproval(decision.approval_id);

Reference pattern · SDK wrappers on the roadmap

1 · Register

Create the agent record so every action carries an internal identity and an accountable owner.

2 · Authorize

Call the authorization endpoint with the action, resource and amount. Honour the returned effect.

3 · Record

Report what the runtime actually did. That result becomes part of the evidence record and its digest.

Integration readiness

  • OpenAI

    Tool-calling agents and assistants

    Reference pattern
  • Anthropic

    Claude agents using tool use

    Reference pattern
  • Google Gemini

    Function-calling agents

    Reference pattern
  • LangChain

    Callback and tool wrappers

    Reference pattern
  • CrewAI

    Multi-agent task authorization

    Planned
  • Microsoft Entra

    Workforce identity mapping

    Planned
  • Model Context Protocol

    Tool-level authorization checks

    Planned
  • Slack

    Approval notifications in channel

    Planned
  • GitHub

    Agent change review evidence

    Planned

What is available today

  • Authorization with recorded decision inputs and policy version.
  • Approval requests with an auditable decision trail.
  • Action evidence with SHA-256 integrity digests.
  • CSV and JSON export for audit and incident review.

Official Python and JavaScript SDK packages are on the roadmap. The HTTP contract is stable enough to integrate against today, and the examples above are reference patterns rather than published packages.